Professional Documents
Culture Documents
6 Popic and Moselhi PDS
6 Popic and Moselhi PDS
Project Delivery Systems Selection for Capital Projects Using the Analytical
Hierarchy Process and the Analytical Network Process
1 2
Zorana POPIC and Osama MOSELHI
1
Department of Building Civil and Environmental Engineering, Concordia
University, 1455 de Maisonneuve Blvd. W, Montreal, Quebec, Canada H3G 1M8;
PH (514) 575-9641; FAX (514) 848-7965; email: z_popic@encs. concordia.ca
2
Department of Building Civil and Environmental Engineering, Concordia
University 1455 de Maisonneuve Blvd. W, Montreal, Quebec, Canada H3G; PH
(514) 848-2424 Ext 3190; FAX (514) 848-7965; email: moselhi@cbs-
engr.concordia.ca
ABSTRACT
In this paper, Analytical Hierarchy Process (AHP) and Analytical Network
Process (ANP) are compared as methods for determining relative weights of factors
in selecting the most suitable project delivery system (PDS) for capital projects. The
AHP considers the elements of each cluster as only affecting the elements of one
other cluster and being affected by elements of one other cluster, whereas the ANP
considers additional dependencies between elements. In selecting a PDS,
interdependencies among factors of different categories exist, therefore ANP is
considered here for its expected suitability. ANP requires additional effort in
constructing a network and additional judgments. A network was developed by
adding dependencies between specific elements to a hierarchy. Both methods were
applied to a case study. ANP generally favored the factors that influenced additional
elements through network connections. In the example analyzed, the overall ranking
of factors by ANP was not consistent with all the pairwise comparisons, which
reveals a limitation of the ANP. This paper augments the research in evaluating the
appropriateness of AHP versus that of ANP in selecting the most suitable project
delivery system. It provides an example of how the priorities of factors by hierarchy
and by network differ for an actual decision problem.
The proposed DSS consists of five tiers. Tier 1 evaluates the use multiple
PDS’s on a project. Tier 2 evaluates whether public-private partnership (PPP) should
be considered for public projects, and screens PPP alternatives. Tier 3 screens for
applicable factors and alternatives outside of PPP. Tier 4 may include one or two
processes of multi-criteria analysis (MCA): Process 1 for non-PPP alternatives, and if
PPP is considered, it also includes Process 2 for PPP alternatives. Processes 1 and 2
rank the available PDS’s in order of suitability. Tier 5 value for money analysis
applies if PPP is considered, and evaluates the most suitable PPP and non-PPP
alternatives. This paper concentrates on the Tier 4, Process 1, which expands on a
method developed by the CII (2003). In addition to the twelve PDS alternatives
included in the CII IR 165-2 (2003), the proposed DSS considers three additional
non-PPP alternatives (Design-Negotiate-Build, Owner-Build and Integrated Project
Delivery) to account for current state of practice.
Selection factors were identified from more than twenty literature sources
some of which rely on expert surveys (CII, 2003) and in depth interviews with
industry participants experienced in using various PDS’s (Touran et al. 2009).
Furthermore, case studies were analyzed to identify PDS selection factors. Ten
factors were identified as Key inputs, which can eliminate some of the PDS
alternatives or some of the selection factors (Tier 3). Six of those factors were also
included in the MCA (Tier 4, Process 1). All the factors considered in MCA were
arranged into a hierarchy, which has five levels, with 67 factors at the lowest level as
shown in Figure 1. The 240 REV’s from the CII IR 165-2 are incorporated in the
proposed DSS, and preliminary values for the three additional PDS’s and 47
additional selection factors (765 REV’s) are proposed, which may be adjusted by the
user as warranted.
selection factor, not changing from project to project. The factors at a lower level of
hierarchy influence a criterion at the level above, to which they are connected (see
Figure 1). Elements of the hierarchy or network inducing the goal and the criteria or
factors on various levels are referred to as nodes. The hierarchy structure was
developed by clustering the factors into categories and subcategories. Factors that
have the greatest influence on a certain objective or condition were grouped together.
Some factors were identified as representing general ideas and others as representing
specific aspects of those general ideas. For example, factor ‘Legal and regulatory
constraints’ represents a general idea or a category, whereas specific types of
constrains represent factors in that category. Furthermore, categories were clustered
together. For example, two categories ‘Legal and regulatory constraints’ and
‘Political considerations’ were grouped into a higher level category ‘Regulatory and
political considerations,’ as they are more similar to one another than to the other
categories, in the way they affect the decision. Seven major categories were identified
at the highest level of hierarchy: Cost, Time, Relationships and Process, Project
Characteristics, Owner Characteristics and Regulatory and Political Considerations.
Per the AHP theory developed by Saaty (1980), the alternatives are evaluated only
with respect to the factors at the lowest level of their respective branch, i.e. the factors
not influenced by any other factor.
Network structure was developed from the hierarchy, by adding dependency
connections between nodes of the hierarchy. The principle in identifying such
connections was in asking: “Does this factor influence or is it influenced by any other
element, aside from those to which it is already connected in the hierarchy.” For
example factors F20 ‘Facilitate early cost estimates’ and F21 ‘Delay or minimize
expenditure rate’ affect cost but also time. Considering all the factors, sets of two or
more factors that may affect an objective in opposite ways were identified, as their
relative importance could be meaningfully compared. For example F4 ‘Lowest cost’
and F26 ‘Optimize lifecycle cost ‘have opposite effects on F6 ‘Quality and
maintainability.’ In developing this network model, the clusters were kept unchanged.
This means that any new connections that could not be used for pairwise comparisons
(i.e. if there were not at least two nodes from the same cluster influencing another
node) were discarded, since the theory of AHP/ANP allows only the nodes within the
same cluster to be compared. The resulting structure is a network as there are
instances where nodes of one cluster influence, or are influenced by nodes of multiple
other clusters, and nodes within a cluster may influence one another. Additional
connections require additional pairwise comparisons, referred to as ‘network
comparisons.’ They affect the priorities of all elements in the decision network. This
forms a simple network as defined in the AHP/ANP theory without sub networks.
The additional ‘network connections’ are shown in Figure 2. The arrows point from
an element that is being influenced to an element that is influencing. Constructing a
network or a hierarchy relies on an understanding of the problem and involves a
certain level of subjectivity inherent in the AHP/ANP method, (Saaty, 1980).
The structure of a model determines which elements will be compared to one
another. Having created a structure, the task is to make all required pairwise
comparisons. Comparison judgments can be expressed linguistically and converted to
numeric values. AHP/ANP theory assigns these values to a matrix that corresponds to
Level Chose PDS
1
2
Cost Time Quality Relationships Project Owner Regulatory and political
and process characteristics characteristics considerations
Risk and Avoid Collaborative Project Scope Owner’s Owner’s Owner’s Legal and
3 Integration Function Change preferences Political
reward claims and process environment definition capabilities goals regulatory
considerations
sharing disputes constrains
the structure of the model. Priorities are obtained by “raising the matrix to arbitrarily
large powers and dividing the sum of each row by the sum of elements in the matrix,”
(Saaty, 1980). Software, such as Super Decisions, automates the AHP/ANP matrix
calculations. The inputs are, first the structure of the hierarchy or network, including
all the elements and their dependency connections, and then the pairwise comparisons
between elements. The outputs are relative priorities of the elements.
Project description
Selection of a PDS for a project encompassing portions of Miami Intermodal
Center (MIC) Core, Phase I was studied by Minchin, Thakkar and Ellis (2007). The
project includes a rental car facility (RCF) building, foundations, underground
utilities and bridge, terminal access roadways, tunnels and bridge, the stations for
two transit systems – Tri-rail (a local rapid transit system) and the MIA Mover
(people mover connecting the MIC with the airport), and MIC/MIA Guideway
Foundation. The estimated cost of this work was $230-$250 million.
The owner, Florida Department of Transportation (FDOT) did not have
experience with all available PDS’s, but was willing to explore a new PDS, even if
that meant overcoming certain regulatory constraints. FDOT had experience with
construction projects but not with all the project types (vertical construction) that this
project incorporated. Aside from FDOT, numerous other public and private parties
had interest in the project and the project funding came from multiple sources.
Elevated fuel distribution centers on every floor of the RCF were to be used for the
first time in the U.S. The appearance and the experience of the public spaces were
very important as this facility would be part of the first impression of Miami and the
U.S. to many visitors.
Construction Research Congress 2014 ©ASCE 2014 1345
The criteria important for PDS selection were owner’s control of design,
ability to meet or exceed schedule requirements, highly qualified contractor, highly
qualified designer, budget/cost control, project team formation and constructability
input to the design (Minchin, Thakkar and Ellis, 2007). In addition, from the
explanation of the PDS selection it can be understood that the ability to overlap
design and construction was valued. The decision maker sought to reduce the risk to
the owner but also to minimize the risk for all parties and to foster non-adversarial
relationships. For this public owner, it was important that all the construction work
would be bid competitively, but that the trade contractors were pre-qualified. A
mechanism to share savings between the owner and the main contractor was
instituted. Ability to handle change and the unexpected was important. (Minchin,
Thakkar and Ellis, 2007)
factors, including the top three factors by ANP, influenced an additional element,
whereas four of those 13 factors were influenced by an additional element. Two
factors had both types of network relationships. By analyzing the changes in ranking
and priorities of factors, having introduced network connections, it was found that 9
out of 14 factors that influenced one additional element directly, and 6 out of 9
factors that influenced only one element, but which element in turn influenced an
additional element through network connections, had higher or equal ranking and
higher priorities in network than in hierarchy. On the account of this, priorities and
rankings of all the remaining factors were lower by network than by hierarchy, except
for one factor that had equal ranking and lower priority by network than by hierarchy.
Four factors that had the greatest percentage gain in priorities accounted for 76% of
the overall gain in priorities and had an average gain in ranking of 17 places. Those
were the factors F72, F20, F26, and F14. Their respective rankings were 24, 13, 21
and 42 by AHP and 3, 1, 6 and 22 by ANP. A set of conditions unique to these four
factors was that each influenced one additional element through its network
connection, was judged as more important than its counterpart in the respective
pairwise comparison, and was not influenced by any additional element.
A measure of consistency of the two methods is to compare the overall
ranking of the factors to their ranking within clusters, which results directly from
pairwise comparisons. It was found that, in the AHP, the ranking order within clusters
was preserved in the overall ranking, whereas in the ANP this was not always the
case. The greatest discrepancy was for cluster ‘Cost.’ The ranking was F16, F40,
Construction Research Congress 2014 ©ASCE 2014 1347
F20, F26 and F21 within the ‘Cost’ cluster and in the overall ranking by AHP,
whereas in the overall ranking by ANP the ranking was F20, F16, F26, F40 and F21.
Therefore, the ANP ranking did not reflect all the judgments expressed, which reveals
its limitation compared to AHP. Table 2 shows that the resulting ranking of the PDS’s
is the same by the two methods, and ratings of PDS’s are very close. Integrated
Project Delivery ranked first, followed by Constructions management at risk and Fast
Track. Since the REV’s did not change, this indicates that the system may have low
sensitivity to factors priorities and that it might favor those highest ranked PDS’s over
the others. Further sensitivity analyses should be performed to explain this result.
CONCLUSION
Proposed DSS for selecting the most suitable project delivery system for
capital projects is presented and illustrated through a case study. AHP and ANP were
compared as methods for determining relative priorities of selection factors. The
network was developed by adding meaningful connections to the hierarchy, based on
the project characteristics. As expected, the ranking and priorities of selection factors
were different for the two methods. The ANP favored those factors that participated
in network connections by influencing additional elements. The factors that were
judged as more important in the additional network comparisons, but were not
influenced by additional elements through network, had the most significant increase
in priorities and ranking by ANP as compared to AHP. Similar effects could be
expected for other network connections, but these findings cannot be generalized
without further studying this and other examples. AHP preserves the ranking of
factors established though pairwise comparisons within clusters. In this example of
ANP, some of the rankings within clusters were overruled in the overall ranking, as a
result of additional network-specific judgments. This indicates that ANP may have a
greater potential for inconsistency. The AHP is simpler to implement without special
software and to integrate with a database of the PDS relative effectiveness values
into a stand-alone application. The resulting ranking of project delivery systems was
the same by both methods; IPD 1st, CMR 2nd and Fast track 3rd. The sensitivity of
the system should be further examined to check whether it is balanced with respect to
various PDS’s. The contribution of this paper is the comparison of the relative
priorities of the same set of factors, by AHP and by ANP, applied to a real decision
problem and in examining the effects of additional network connections on the
priorities and ranking of factors. Future work should examine the impact of
constructing other networks for the same project while using the same set of factors
in the selection criteria, and study other similar examples
REFERENCES
Al Khalil, M. (2002). “Selecting the appropriate project delivery method using AHP”,
International Journal of Project Management, Vol. 20, Issue 6, pp. 469–474.
Almazroa, D. (2003). “Project delivery system decision framework using the
weighting factors and analytic aierarchy rrocess methods.” PhD thesis, School
of Engineering, University of Pittsburgh. Pennsylvania, USA.
Anderson, S. and Oyetunji, A. (2003). “Selection procedure for project delivery and
contract strategy”, Proc., Construction Research Congress: Wind of Change:
Construction Research Congress 2014 ©ASCE 2014 1348